yolov8CIoU Loss替代GIOULoss代码

### YOLOv8 使用 CIoU Loss 替代 GIoU Loss 的代码实现 在目标检测任务中，CIoU Loss 是一种改进的损失函数，相较于 GIoU Loss，它考虑了边界框之间的重叠区域、中心点距离以及长宽比的影响。以下是基于 PyTorch 的 YOLOv8 实现中将 CIoU Loss 替代 GIoU Loss 的代码示例。 #### 1. 损失函数定义 CIoU Loss 的核心公式如下： \[ CIoU = IoU - \frac{d^2}{c^2} - \alpha v \] 其中： - \(d\) 是预测框与真实框中心点的距离。 - \(c\) 是包含两个框的最小外接矩形的对角线长度。 - \(v\) 表示长宽比的一致性：\(v = \frac{(ar_{gt} - ar_{pred})^2}{\kappa}\)。 - \(\alpha\) 是一个平衡因子，用于调整 \(v\) 的权重[^1]。 以下是 CIoU Loss 的实现代码： ```python import torch def bbox_iou(box1, box2, x1y1x2y2=True, GIoU=False, DIoU=False, CIoU=True, eps=1e-7): # Returns the IoU of box1 to box2. box1 is 4, box2 is nx4 box2 = box2.T # Get the coordinates of bounding boxes if x1y1x2y2: # x1, y1, x2, y2 = box1 b1_x1, b1_y1, b1_x2, b1_y2 = box1[0], box1[1], box1[2], box1[3] b2_x1, b2_y1, b2_x2, b2_y2 = box2[0], box2[1], box2[2], box2[3] else: # transform from xywh to xyxy b1_x1, b1_x2 = box1[0] - box1[2] / 2, box1[0] + box1[2] / 2 b1_y1, b1_y2 = box1[1] - box1[3] / 2, box1[1] + box1[3] / 2 b2_x1, b2_x2 = box2[0] - box2[2] / 2, box2[0] + box2[2] / 2 b2_y1, b2_y2 = box2[1] - box2[3] / 2, box2[1] + box2[3] / 2 # Intersection area inter = (torch.min(b1_x2, b2_x2) - torch.max(b1_x1, b2_x1)).clamp(0) * \ (torch.min(b1_y2, b2_y2) - torch.max(b1_y1, b2_y1)).clamp(0) # Union Area w1, h1 = b1_x2 - b1_x1, b1_y2 - b1_y1 + eps w2, h2 = b2_x2 - b2_x1, b2_y2 - b2_y1 + eps union = w1 * h1 + w2 * h2 - inter + eps iou = inter / union if CIoU or DIoU: cw = torch.max(b1_x2, b2_x2) - torch.min(b1_x1, b2_x1) # convex (smallest enclosing box) width ch = torch.max(b1_y2, b2_y2) - torch.min(b1_y1, b2_y1) # convex height if CIoU: # https://github.com/Zzh-tju/DIoU-SSD-pytorch/blob/master/utils/box/box_utils.py#L47 c2 = cw**2 + ch**2 + eps # convex diagonal squared rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2)**2 + (b2_y1 + b2_y2 - b1_y1 - b1_y2)**2) / 4 # center distance ** 2 v = (4 / math.pi**2) * torch.pow(torch.atan(w2 / (h2 + eps)) - torch.atan(w1 / (h1 + eps)), 2) with torch.no_grad(): alpha = v / (v - iou + (1 + eps)) return iou - (rho2 / c2 + v * alpha) # CIoU elif DIoU: c2 = cw**2 + ch**2 + eps # convex diagonal squared rho2 = ((b2_x1 + b2_x2 - b1_x1 - b1_x2)**2 + (b2_y1 + b2_y2 - b1_y1 - b1_y2)**2) / 4 # center distance ** 2 return iou - rho2 / c2 # DIoU return iou # IoU ``` #### 2. 替换 YOLOv8 中的损失函数 YOLOv8 的训练流程通常会定义一个损失函数模块。假设当前使用的是 GIoU Loss，可以将其替换为 CIoU Loss。以下是修改后的代码片段： ```python class ComputeLoss: def __init__(self, model): self.model = model self.hyp = model.hyp self.device = next(model.parameters()).device def __call__(self, preds, targets): loss_box = torch.zeros(1, device=self.device) for pred in preds: # 计算边界框损失 box_loss = self.bbox_loss(pred[:, :4], targets[:, :4]) loss_box += box_loss return loss_box def bbox_loss(self, pred_boxes, target_boxes): return 1 - bbox_iou(pred_boxes.T, target_boxes, x1y1x2y2=False, CIoU=True) ``` #### 3. 训练过程中的调用 在训练过程中，确保 `ComputeLoss` 类被正确初始化并调用。例如： ```python loss_fn = ComputeLoss(model) total_loss = loss_fn(preds, targets) total_loss.backward() optimizer.step() ``` 通过上述代码，YOLOv8 的边界框回归损失从 GIoU Loss 替换为了 CIoU Loss。 ---